InsForge bets coding agents, not humans, are the backend's real users

InsForge is an open-source backend platform designed for AI coding agents rather than humans. It bundles Postgres, auth, S3-compatible storage, edge functions, an OpenAI-compatible model gateway, and hosting behind one CLI and MCP toolset, so a coding agent can deploy, operate, debug, and scale a full-stack app end to end without a human touching a dashboard.

The founding bet came from experience: co-founders Hang Huang (ex-Amazon PM) and Tony Chang (ex-Databricks) found that MCP servers for Supabase, Vercel, and GitHub bloated their agents' context windows with 10K+ token payloads while still hiding the state agents need - telemetry, configs, logs, runtime behavior. They concluded that infrastructure built for humans was the bottleneck and rebuilt the stack around agents as primary operators, with structured outputs, scoped permissions, and reversible writes.

In March 2026 InsForge joined YC's P26 batch and launched version 2.0: the company reports #1 Product of the Day on Product Hunt, GitHub's #1 trending repository, and GitHub stars growing from 1,500 to 6,600+ in one month. The project is Apache 2.0 and self-hostable via Docker Compose with parity to the cloud tier, a deliberate hedge against lock-in for teams that want an exit route if the company disappears.

By August 2026 the repo sat at roughly 12.7k GitHub stars, with a release cadence of three feature releases in under three weeks. Independent coverage praised the MCP-first architecture but flagged the open questions: security guidance drifting from a fast-moving codebase, RLS policies written by agents, and whether the Compute tier for long-running containers will graduate from private preview.